Chain-of-Thought

Two answering machines play a game of telephone, mishearing and misremembering as they repeat one another’s stories. Their conversation spirals and the stories become increasingly hallucinatory with each telling. What sounds at first like a memory begins to blur into a senseless chain of next likely words.

Model Collapse creates a closed generative circuit that starts poisoning itself. The answering machines, too, carry the symptoms of their condition: each is milled from burl wood—tumorous knots that form when a tree is infected and grows around its wound. The language generated by the answering machines has been trained on the largest public collection of dreams as well as the entire body of emails from the American energy conglomerate Enron that collapsed spectacularly in 2001. The largest bankruptcy in US history, the company unraveled because its fraudulent accounting was based on anticipated and predicted profits, not its current income. Projection and speculation ran the numbers until they stopped making sense. During Enron’s trial, its thousands of internal emails entered the public domain and became one of the foundational datasets for the large language models that are hollowing out office and cognitive work today.

Cooper Jacoby draws eerie parallels between different systems: failed business models which are compressed down into AI training data; AI models which collapse when repeatedly trained only on their own output; and cows as living models of how markets parcel out bodies into functional components. Each one collapses when it becomes closed in on itself. Mad Cow Disease infects the brain of cows when they ingest “beef” – aka themselves. Equally, large language models start failing when they’re only fed themselves – they begin to hallucinate, forget, and average out into slop.

“Chain-of-thought” traces these parallel dynamics across four bodies of work that reanimate obsolescing hardware. Accompanying Model Collapse, there are Twisted Pairs, a series of sculptures that resemble wall-mounted telephones, made from cast cow ears nestled into each other with a cord of corroded, barbed wire. As America’s “Last Frontier” was colonized in the 19th century, private cattle fencing became transmissions lines for the spread of telecommunications through rudimentary local networks, a precursor to the way data centers now graft onto power-plants. Every now and then Twisted Pairs emit a whistle to one another, like lost signals from a severed grid.

Mechanizing another part of the cow, Domesticator combines flatbed scanners with silicone cow tongues that wag slowly back and forth as they follow dice moving beneath the scanner’s glass. The scanner and its Optical Character Recognition software were pivotal in expanding the archive of digitized text, translating print into machine-predicted number strings. Domesticator maps this loop from language to prediction and back again: moving from tongue to page to code and back to speech again.

The fourth body of work, Mutual Life is a series of clocks in which teeth, both human and animal, slowly circle the surface of mirrored stainless-steel domes. Each work is a portrait of a different person’s biological age, as opposed to their chronological age, moving faster or slower depending on the individuals supposed cellular age compared to population averages. Insurance companies are increasingly interested in this biological age to more accurately predict and price lifespan – another step of turning life itself into a quantifiable asset that can be predicted and hedged against.

From rusted cattle fencing to corporate fraud training data, from cow tongues to answering machines, “Chain-of-thought” roots an increasingly hallucinatory world in the machines that digest us. However, these systems require our words and attention to survive; without us, they stare in the mirror and start devouring themselves.

The artist thanks Mayra Berrones, Agnes Fury Cameron, Leo Elia, Juergen and Hildegard Findeisen, Robbie Fitzpatrick and Fitzpatrick Gallery, Aric Grauke and Stephanie Seidel.